Longitudinal monitoring of circulating tumor DNA to detect relapse early and predict outcome in early breast cancer

Abstract

Purpose

Detection of molecular residual disease (MRD) allows for the identification of breast cancer patients at high-risk of recurrence, with the potential that early initiation of treatment at early stages of relapse could improve patient outcomes. The Invitae Personalized Cancer Monitoring™ assay (PCM) is a newly developed next-generation sequencing approach that utilizes up to 50 patient-specific, tumor-informed DNA variants, to detect circulating tumor DNA (ctDNA). The ability of the PCM assay to detect MRD before clinical relapse was evaluated.

Methods

The cohort included 61 female patients with high-risk breast cancer who underwent neoadjuvant chemotherapy. Plasma samples were collected before and during neoadjuvant therapy, after surgery and during monitoring. PCM was used to detect ctDNA at each time point.

Results

The sensitivity to detect ctDNA in plasma from patients who relapsed during the monitoring phase was 76.9% (10/13). Specificity and positive predictive values were both 100% with all (10/61, 16%) of the patients who had ctDNA detected during the monitoring phase subsequently relapsing. Detection of ctDNA during monitoring was associated with a high-risk of future relapse (HR 37.2, 95% CI 10.5–131.9, p < 0.0001), with a median lead-time from ctDNA detection to clinical relapse of 11.7 months.

Conclusion

PCM detected ctDNA in patients who relapsed with a long lead-time over clinical relapse, shows strong association with relapse-free survival and may be used to identify patients at high-risk for relapse, allowing for earlier intervention.

Supplementary Information

The online version contains supplementary material available at 10.1007/s10549-024-07508-2.

Introduction

Despite advancements in treatment for breast cancer over the past three decades, the risk for recurrence remains, with approximately 15% of patients destined to relapse [1, 2]. For patients with the most common cancer subtype- hormone receptor positive/HER2 negative—the risk of relapse, including distant metastasis, remains steady from 5 to ≥ 20 years post-diagnosis, with over half of distant metastasis occurring > 5 years post-diagnosis for patients [3, 4]. Currently, guidelines for post-treatment management in the absence of clinical symptoms include mammography every 12 months with no routine laboratory or imaging studies to monitor recurrence [5]. Because localized and regional recurrence have been associated with a higher risk of developing distant metastasis and decreased survival [6–10], novel strategies to monitor and detect recurrence at the earliest stages are critical in improving outcomes for patients with recurrent breast cancer.

Cancer cells that remain after curative treatment (molecular residual disease, MRD) serve as a source for both locoregional and distant recurrence [11]. The ability to detect MRD would allow patients at high-risk for recurrence to be identified, which may translate to improved outcomes by initiating treatment at the early stages of the disease process [12–14]. MRD cannot be detected by imaging or clinical examination; rather, molecular technologies have been developed that identify the presence of circulating tumor DNA (ctDNA) by detecting mutations unique to tumor cells within blood specimens. As ctDNA may represent as little as < 0.01–0.1% of the total cell-free plasma DNA (cfDNA) concentration in patients with early-stage breast cancer, methods of detection with high sensitivity and specificity are required to accurately detect and distinguish low levels of ctDNA from the majority cfDNA shed from normal cells throughout the body [15].

Advances in next-generation sequencing (NGS) technology, such as the use of unique molecular identifiers and multiplex PCR, allow for patient-specific assays to detect ctDNA with high sensitivity [16]. For example, tracking ~ 500 patient-specific DNA variants demonstrated higher sensitivity than digital PCR (dPCR) in detecting ctDNA, with a sensitivity of 81% in patients with recently diagnosed metastatic breast cancer [17]. This approach, however, proved challenging as most breast cancer patients in the study had far fewer variants (median 57) than those used in the benchmark assay. In contrast, in a study using a 1–7 variant assay, the sensitivity to detect MRD was 31% [18]. Two studies have, however, utilized sequencing panels with 16–51 variants, resulting in sensitivities of 86–89% and lead-times of molecular recurrence compared to clinical detection of 9–12 months [19, 20].

A number of studies have evaluated the ability to detect MRD in patients with early breast cancer during the monitoring phase, with sensitivities ranging from to 79–93% and median lead-times of 7.9–11 months [19, 21–23]. These studies used a variety of approaches including digital PCR and NGS-methodologies. Invitae Personalized Cancer Monitoring™ (PCM) is a pan-cancer, tumor-informed, long-term monitoring assay that provides personalized ctDNA detection using patient-specific panels (PSPs) of 18–50 somatic variants from the peripheral blood of patients with solid tumors [24]. This assay has recently been shown to be effective in monitoring residual or recurrent disease in patients with early-stage non-small cell lung cancer (NSCLC) [25]. In our current study, we evaluated the ability of PCM to detect MRD in patients with early-stage breast cancer.

Materials and methods

Patient cohort

In this proof-of-principle study, ctDNA levels were monitored retrospectively in plasma samples from patients with early breast cancer who enrolled in the ChemoNEAR sample collection study (Research Ethics Committee ref. no. 11/EE/0063) between November 2012 and September 2016. Sixty-one patients who had WES data sufficient to generate PSPs had multiple plasma samples collected during the monitoring phase that were available for research purposes. The study was approved by the East of England Health Research Authority ethics committee. Written informed consent was obtained from all participants. All patients had primary breast cancer without evidence of distant metastatic disease. Patients had pre-, during, and post-treatment blood samples collected for ctDNA analysis. Hormone receptor (HR) and HER2 status were determined from core biopsies taken at the time of diagnosis and tumors were classified as HR positive/HER2 negative (HR +), HER2 positive regardless of HR status (HER2 +), triple-negative (TNBC) or unknown. All patients underwent neoadjuvant chemotherapy (NAC) with or without adjuvant therapy, as previously described [22].

Sample collection and processing

Processing of blood and tumor specimens and isolation of plasma DNA was performed as previously described [23]. Plasma sample collection times included before initiation of NAC (“baseline”), during the second cycle of NAC, ≤ 8 weeks after surgery (“landmark”) and every three months for the first two years and every six months thereafter for the monitoring samples.

Library preparation, sequencing and ctDNA detection

NGS libraries from tumor and germline DNA were sequenced as previously described [24]. Variants identified in the germline specimens including clonal hematopoiesis of indeterminate potential (CHIP) variants were subtracted from the variants detected in the tumor. Patient-specific panels (PSPs) targeting tumor-specific variants were designed using proprietary algorithms (Invitae Corporation, San Francisco, CA). Sequencing data were processed and variant detection was performed as previously described, with a p-value of < 0.001 used to determine whether ctDNA was absent [25]. Briefly, ctDNA detection was determined using a one-sided Poisson test assessing whether targeted variants were observed more often than would be expected based on a sample-specific noise model. The null hypothesis was that there was no difference between the observed and expected variants. The null hypothesis was rejected if the p-value was < 0.001 for a sample. The ctDNA status of absent, present or indeterminate were determined based on whether the null hypothesis was true, false or QC failure were observed, respectively.

Statistical analysis

Sensitivity was defined as a percentage of patients with a relapse, distant or local, who had ctDNA detected prior to relapse. For this study, patients who did not have plasma samples collected within the 6 months prior to relapse were excluded from the total number of patients who relapsed. Specificity was similarly evaluated in the subset of patients who were relapse-free at the time of follow-up, it is noted that specificity may be underestimated as patients may relapse in the future. Associations between baseline and second cycle ctDNA status and relapse was determined using an exact pooled Z test. Time-to-relapse (TTR) analysis was performed in R version 4.3.2 using the “survival” package version 3.5–7 [26]. TTR was used as the disease progression endpoint. Because ctDNA detection during the monitoring phase is a time-dependent covariate, data were fit to a Cox proportional hazards model rather than using a log-rank test. Baseline ctDNA detection was modeled as a constant covariate. The p-values for the Cox models were calculated using the likelihood ratio test. Patient 20,000,006,307 had a missing surgery date and was thus excluded from RFS analysis. Lead-time was calculated as the time from the earliest MRD positive plasma sample post-surgery until relapse, for all patients where MRD was detected prior to relapse. Differences in lead-time between subtypes were evaluated using the Kruskal–Wallis test.

The Sankey Plot describing the progression of ctDNA status and relapse was generated using the “plotly” package version 5.18.0 for Python 3. Post-surgery samples were binned into increments based on time post-surgery (the “6 mo” bar corresponds to samples from 0 to 6 months post-surgery). A patient was considered to be perpetually ctDNA + after the first post-surgery positive result. Similarly, once a patient entered one of the “Relapsed” categories, they were considered perpetually relapsed. If a patient had no samples available for a given bin, they were assumed to maintain their previous ctDNA status, if any. Individual samples with an indeterminate ctDNA result were excluded from this analysis.

Results

Patient characteristics

Clinical characteristics of the 61 female patients included in this study are shown in Table 1. Tumor types included HR + (34%), HER2 + (33%), TNBC (26%) and unknown (6%). Two patients developed brain-only metastasis (patients 20,000,004,708 and 20,000,005,606). All patients received standard neoadjuvant therapy. Of note, patient 20,000,005,606, classified as HR + /HER2- at baseline, had a surgical specimen that was HR + /HER2 + after completion of NAC. Date of surgery was missing for one patient (20,000,006,307). Nineteen patients had pathological complete response (pCR), including two patients who later developed distant metastasis. Overall, at 55 months median follow-up (range 21–77), 13/61 (21%) patients had relapsed. All patients provided at least two plasma samples (range 2–12, median 8) during the monitoring phase.

Table 1.

Clinical characteristics of 61 patients evaluated for ctDNA using PCM. Data shown as N (%)

Patient characteristicsa	All (n = 61)
Median Age at Diagnosis (years)	51
Range (years)	28–75
Menopausal status
Pre	30 (49)
Peri	1 (2)
Post	24 (39)
Unknown	6 (10)
Tumor size
T1	14 (23)
T2	32 (52)
T3	12 (20)
Unknown	3 (5)
Lymph node status
Positive	34 (56)
Negative	24 (39)
Unknown	3 (5)
Subtype
HR +	21 (34)
HER2 + b	20 (33)
TNBC	16 (26)
Unknown	4 (7)
pCR
Yes	19 (31)
No	42 (69)
Relapse
Yes	13 (21)
No	48 (79)
Lead-time (months, median)	11.7
Range	3.9–58.9
PSP size
18–50	49 (80)
11–17	12 (20)
Baseline ctDNA positive	38 (68) N = 56
Monitoring ctDNA positivec	11 (18)

^aNo association was found between ctDNA detection during monitoring and any of the available clinical and pathological features (data not shown)

^bOne patient had HER2 negative disease based on evaluation of the core needle biopsy and did not receive trastuzumab during NAC. The surgical resection was HER2 positive and the patient was treated with adjuvant trastuzumab. This patient later developed brain metastasis

^cOne patient had ctDNA detected after relapse. The patient relapsed with bone and lung cancer. The last plasma sample before relapse was collected > 17 months before metastases were clinically detected

Characteristics of PCM

The median number of variants included in the PCM assays was 33 (range 11–50). Twelve patients had a low number of somatic variants detected by whole exome sequencing, resulting in PSPs with 11–17 variants, which is smaller than the standard PCM PSP size of 18–50 variants. Each subtype had tumors that required the creation of PSPs with 11–17 variants (HR + 5/21, 24%; HER2 + 6/20, 30%; TNBC 1/16, 6%, Supplementary Fig. 1).

Baseline ctDNA

Baseline plasma samples were available for 56/61 (92%) patients; of these, 36 had an additional plasma sample taken at the time of a second cycle of NAC. At baseline, ctDNA was detected in 38/56 cases (67.8%); 5 cases were indeterminate, and 13 cases were negative (Fig. 1). Detection of ctDNA at baseline was highest in patients with TNBC (14/15, 93.3%) followed by HR + (11/18, 61.1%) and HER2 + (10/19, 52.6%) tumors. There was an association between ctDNA detection at baseline and larger tumor size (Supplementary Fig. 2); however, no significant associations between ctDNA and any other clinical or pathological characteristics were detected. The rates of pCR were similar in those with and without ctDNA detection at both baseline (14/38 (36.8%) vs. ctDNA negative 3/13 (23.1%), p-value = 0.18) and at second cycle of NAC (4/13 (30.8%) vs. 6/22 (27.3%), p = 0.41). ctDNA was detected at baseline for both patients who had pCR but later relapsed. None of the patients (0/13) who were ctDNA negative at baseline relapsed; in comparison, 10/38 (26.3%) patients with ctDNA detected at baseline developed distant metastasis (p = 0.02) (Supplementary Fig. 3).

Fig. 1.

Baseline characteristics a Proportion of baseline samples per subtypes (center) and ctDNA detection status (present, absent and indeterminate) for HR + (blue), HER2 + (pink), TNBC (green) and unknown (orange) subtypes; b Heatmap showing the clinical characteristics of the 56 patients with baseline ctDNA characterization. MRD call (Positive =  + , Negative = -, Indeterminate = Ind.) at baseline is represented along with log₁₀ Max detected allele frequency (AF) in the plasma. Patients are grouped per subtype and sorted by Max AF

Landmark and monitoring ctDNA

None of the landmark plasma samples taken after neoadjuvant chemotherapy and within 8 weeks of surgery (n = 46) were ctDNA positive (Fig. 2); in contrast, during monitoring, 10/61 (16.4%) patients became ctDNA positive and all 10 subsequently relapsed (Fig. 3). The average time to first detection of ctDNA in the monitoring samples was 16.7 months post-surgery. Detection of ctDNA during monitoring was associated with a high-risk of future relapse (HR 37.2, 95% CI 10.5–131.9, p < 0.0001), with a median lead-time from ctDNA detection to clinical relapse of 11.7 months (range 3.9–58.9) (Fig. 4). Median time from surgery to first MRD detection was 15 months and 25, 17 and 6 months for HER2 + , HR + and TNBC, respectively. One patient had ctDNA detected after relapse; this patient was ctDNA positive at baseline, however, monitoring plasma samples were collected more than one year before relapse. Two other patients relapsed (to bone and liver and to adnexal mass) without ctDNA detection during the monitoring phase; at baseline, one was ctDNA positive and the other had an indeterminate call, with no plasma samples collected three and eight months before relapse, respectively.

Fig. 2.

Swimmer plot representing longitudinal ctDNA tracking for all patients in the study For each patient, baseline and monitoring ctDNA results are represented as well as patients characteristics (Age at diagnostic, menopausal status, pre-chemotherapy tumor grade, pCR, panel size and subtype at diagnostic). Of note, for patient 20,000,004,710 ctDNA was detected shortly before relapse, with the symbol for relapse (red X) partially obscuring the symbol for ctDNA detection (green diamond)

Fig. 3.

Sankey plot of disease progression and its relationship to ctDNA detection over time in the patient cohort Baseline samples are grouped together, then additional samples are grouped into 6 month bins, labeled by the last time point post-surgery in that bin. All patients that convert from ctDNA negative (blue) to ctDNA positive (yellow) eventually relapse (maroon), indicating a 100% positive predictive value. Patient 20,000,006,307 was not included due to a missing surgery date

Fig. 4.

Proportional hazards model of time to relapse for all patients participating in the study ctDNA status during monitoring was treated as a time-dependent covariate. The hazard for relapse of ctDNA + was significantly higher than for ctDNA- patients (HR = 37.161, likelihood ratio p-value = 2.84e–09.). Patient 20,000,006,307 was not included due to a missing surgery date

While overall assay sensitivity was 76.9%, when considering only patients who had blood samples within 6 months of clinical relapse, assay sensitivity was 90.9% (10/11), with sensitivities of 100%, 100% and 75% for patients with HR + , HER2 + and TNBC tumors, respectively. The specificity for relapse and positive predictive values were both 100%. The negative predictive value was 100% for HR + and HER2 + and 92% for TBNC subtypes. Patients with HR + tumors had the highest rates of MRD positivity during monitoring (6/21, 28.6%) followed by TNBC (3/16, 18.8%) and HER2 + (1/20, 5.0%) (Fig. 2).

Of note, two patients (HER2 + n = 1, HR + n = 1) that relapsed had brain metastases with no additional known sites of metastasis. For both patients, ctDNA was detected both at baseline and before relapse. Lead-times for these patients were 5.6 and 3.8 months.

Small PSP size

Of the 12 patients with smaller PSPs (11–17 tracked variants), 10 had a baseline sample. Four patients (40%) had ctDNA detected at baseline, two were indeterminate and four were negative. In contrast, when using PSPs ≥ 18 tracked variants, 34/46 (73.9%) patients had ctDNA detected at baseline. Of note, larger panel size did not guarantee detection as 4/9 (44.4%) patients with no ctDNA detected had PSP size of 50 variants. During the landmark and the monitoring periods, none of the patients with panels targeting fewer than 18 variants relapsed or had ctDNA detected (Supplementary Fig. 1).

Discussion

Both the American Society of Clinical Oncology and the National Comprehensive Cancer Network strongly recommend against routine imaging to detect metastatic disease [27]. Thus, alternate strategies to identify patients who are likely to recur are needed to identify patients with MRD so that these high-risk patients can undergo adjuvant therapies, while reducing rates of overtreatment of low-risk patients by sparing them from the possible toxicities and costs associated with additional treatment. Here, we demonstrate that the PCM assay demonstrated a clinical sensitivity of 76.9% for detecting MRD during the monitoring phase in patients who relapsed, with a specificity of 100% and a median lead-time of 11.7 months.

Previous studies have reported on the ability to detect post-treatment MRD, however, these studies utilized different technologies [21–23], included only patients with TNBC [28, 29] or monitored late recurrence (> 5 years post-diagnosis) MRD [20] Two studies also used a tumor-specific approach to detect MRD in cohorts of 49 and 156 patients with heterogeneous subtypes of breast cancer, achieving sensitivities of 88.2–88.9% [19, 30]; unlike our study, however, these two studies were not limited to early breast cancer patients but rather included 20% and 33% of patients having Stage IIIB/C breast tumors. Data presented here represents the first report of the efficacy of using a limited number of tumor-specific variants to detect MRD in early breast cancer patients across all subtypes.

Of note, the PCM assay detected ctDNA in the two patients who had pCR but later relapsed. Patients with high-stage disease at diagnosis remain at risk of relapse despite achieving a pCR [31] and the ability to detect molecular relapse in those individuals who would otherwise be expected to have good prognosis may allow intervention to prevent relapse. In addition, for two patients with HR + tumors, ctDNA was detected while the patients were on adjuvant endocrine therapy. A recent study of patients with colorectal cancer found that in patients who were ctDNA positive four weeks post-surgery, those who received adjuvant chemotherapy were significantly more likely to have ctDNA clearance by week 24 than those who did not receive chemotherapy [32]. Those patients who converted from ctDNA positive to negative had improved disease-free survival compared to those who remained positive or converted from negative to positive. In our study, the PCM assay also detected ctDNA in both patients who developed isolated brain metastases, an average of 4.8 months before clinical detection. This result was surprising given that the blood brain barrier is thought to limit ctDNA from passing into the circulatory system, and a previous study of 55 women with early breast cancer failed to detect ctDNA at any time point for three patients with brain metastasis [22]. While PCM detected MRD in both patients with isolated brain metastases in our study, the number of patients was small. Future studies are thus needed to determine the efficacy of the PCM in detecting brain metastasis.

The timing and number of samples collected during monitoring may affect the ability to detect MRD. In a study that included 55 patients from the same ChemoNEAR study as the patients in our study, 69% of baseline samples and 19% of landmark samples, collected 2–4 weeks post-surgery, were ctDNA positive [22]. In our study, the baseline ctDNA detection rate was similar (67.8%), however, none of the landmark plasma samples, collected up to 8 weeks post-surgery, were ctDNA positive. Of note, in the study by Garcia-Murillas et al. ctDNA detection rates in patients who relapsed were higher (80%) in monitoring samples compared to landmark samples [22]. With 76.9% of patients who relapsed having ctDNA detected in monitoring samples, these data highlight the importance of using PCM in serial sampling after treatment.

The optimal size PSP to effectively detect MRD varies widely: one previous study found that use of 1–7 variants resulted in a sensitivity of only 31% [18] while a second study achieved a sensitivity of 81% attained when ~ 500 variants were queried but this dropped to 19% when a median of 57 variants were included in the PSP [17]. The ability to detect ctDNA with a minimal set of variants is critical for patients with breast cancer who frequently have lower numbers of somatic variants than other cancer types [33]. A previous validation study demonstrated that when PCM included 18–50 variants in the PSP achieved a specificity and sensitivity of > 99.9% [24]. In this study smaller panels with 11–17 variants were required for 12 patients with low numbers of somatic variants.. Smaller PSPs could detect ctDNA in pre-treatment plasma samples although none of the monitoring phase samples had ctDNA detected. Because patients with tumors with low mutational burden may be less likely to relapse, larger studies are needed to determine the efficacy and utility of PCM in patients with lower numbers of tumor-informed DNA variants.

This study has several limitations. Sample size was limited to 61 patients with heterogeneous breast cancer subtypes. Some data points were not available for the baseline and/or landmark period due to missing collection and/or sample processing failure. Patterns of ctDNA detection and sensitivity to detect relapse by subtype differed in our study compared to previous publications: for example, rates of ctDNA detection at baseline were 72% and 50% in patients with HR + and HER2 + tumors, respectively, compared to 48% and 84% in the study from Magbanua et al. [34] and sensitivity was 67% in patients with TNBC in our study compared to 100% in the study by Coombes et al. [19]. In conjunction, the median follow-up time in this study was 55 months; although the HR + group had the highest rates of relapse in this study, this group of tumors generally has a longer time to recurrence. For example, a study of women with high-risk HR + /HER2- breast cancer > 5 years after diagnosis found that 7.2% of patients, all of whom were MRD positive, developed distant metastasis [20]. Thus, additional studies are needed to determine how long-term ctDNA monitoring predicts relapse, especially in HR + /HER2- breast cancer. A number of clinical trials, such as TRAK-ER (ClinicalTrials.gov identifier: NCT04985266), which will determine whether palbociclib and fulvestrant, can defer or prevent relapse in patients with MRD, are underway.

In conclusion, MRD was detected in all breast cancer subtypes with long lead-times over clinical relapse. ctDNA detection during the monitoring phase was associated with significantly shorter relapse-free survival. The PCM assay is, therefore, an effective tool in identifying patients at high-risk for relapse, which will allow for earlier intervention and may improve patient outcomes.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

This work was conceptualized and designed by Isaac Garcia-Murillas, Robert Daber, W. Michael Korn, and Nicholas C Turner. Patient samples and data were collected by Rosalind J Cutts, Giselle Walsh-Crestani, Edward Phillips, Sarah Hrebien, Kathryn Dunne, Kally Sidhu, Stephen RD Johnston, Alistair Ring, Simon Russell, Abigail Evans, Anthony Skene, Duncan Wheatley, Ian E Smith. Data was generated and analyzed by Benjamin Hubert, Chiharu Graybill, Peter M. DeFord, David J. Wooten and Jianhua Zhao. The first draft of the manuscript was written by Rachel E. Ellsworth and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding

This research was funded by Invitae Corp. and Program Funding to the Breast Cancer Now Toby Robins Research Centre (Grant Number CTR-Q5-Y3, Le Cure and RM Charity, NIHR funding to the Royal Marsden and Institute of Cancer Research Biomedical Research Centre supported this work.

Data availability

When not prohibited by patient permissions or privacy laws, the de-identified individual data that underlie the results reported in this article will be made available to researchers. Researchers will be asked to submit a short proposal outlining objectives, research question and analytical methods, and submit IRB approval or determination of exempt status or nonhuman subjects research. For more information on how to access the Invitae data please contact the corresponding author.

Declarations

Competing interests

Robert Daber, Benjamin Hubert, Peter DeFord, David Wooten, Jianhua Zhao, and Rachel E. Ellsworth are current employees of Labcorp Genetics, Inc. Robert Daber, Benjamin Hubert, Chiharu Graybill, Peter DeFord, David Wooten, Jianhua Zhao, Rachel E. Ellsworth and W. Michael Korn are former employees and shareholders of Invitae Corp.

Ethical approval

This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Research Ethics Committee ref. no. 11/EE/0063.

Consent to participate

Written informed consent was obtained from all participants.